Feature Selection Method for Network Intrusion Based on Fast Attribute Reduction of Rough Set

Abstract

Aiming at the problem that independent and redundancy attributes cause classification algorithms' low detection speed and detection rate in network intrusion detection. Hence, a novel feature selection approach for network intrusion based on fast attribute reduction of rough set is proposed in the paper. First, the approach removes independent attributes according to normalized mutual information between condition attributes and decision attributes, then an improved formula for measuring attribute importance based on positive region of rough set is presented. Finally, a fast and recursive attribute reduction method is designed to realize feature selection of network intrusion. KDDCUP1999 data-set are used to experiment. The experimental result shows that compared with similar algorithms, the approach is more effective and efficient in discarding independent and redundancy attributes and in improving intrusion detection performance of classification algorithm.